Yes, you should check normality of errors AFTER modeling. In linear regression, errors are assumed to follow a normal distribution with a mean of zero. Let's do some simulations and see how normality influences analysis results and see what could be consequences of normality violation.
Hope this helps:)